Based on chat with Judith about direct-beam iteration:
- NaN-regions should probably be masked, or else we will get NaN in the workflow result. Should we mask the direct beam file, or should the reduction workflow perform its own masking?
- Should NaN-regions be "filled" with approximate values based on non-NaN neighbours?
Based on chat with Judith about direct-beam iteration: